A Machine Learning-Guided Metaheuristic Framework for the Multidimensional Knapsack Problem

Abstract

This study utilizes the binary version of the Coati Optimization Algorithm (BinCOA), as well as an enhanced version of the Guided Genetic Algorithm (GGA) to find close- to-optimal solutions for the Multidimensional Knapsack Problem (MKP). Our approach leverages the effectiveness and consistency of the BinCOA which is a binary variant of the novel metaheuristic “Coati Optimization Algorithm” – originally used to solve the Knapsack Problem (KP) – to solve the much harder MKP. In addition, we propose an improvement to the GGA by adding a neighborhood local search method which intelli- gently explores nearby solutions. Inspiration is taken from Machine Learning Algorithms, where Q-learning agents are leveraged to repair solutions in the BinCOA and perform local search for GGA. Experimental results show that the BinCOA implementation and improved GGA deliver comparable results to state-of-the-art algorithms and sometimes surpassing them for the Chu&Beasly and Sac-94 OR benchmarks.